AI Supported Forensic Accounting in High Risk Audit Engagements

Abstract

This research introduces a novel, hybrid artificial intelligence framework designed to augment forensic accounting procedures within high-risk audit engagements. Moving beyond conventional data analytics, the proposed methodology
synergistically integrates a rule-based expert system, a neural network anomaly
detector, and a natural language processing module for narrative analysis of unstructured data. The framework is specifically architected to identify complex,
multi-layered financial fraud schemes—such as circular transactions, shell company
networks, and earnings management through related-party dealings—that often
evade traditional audit techniques. A core innovation lies in its application of
swarm intelligence algorithms, inspired by ant colony optimization, to trace the
flow of funds across intricate corporate structures, mimicking the deductive reasoning of a seasoned forensic investigator. We developed and tested the framework
using a proprietary, anonymized dataset comprising 127 historical high-risk audit
cases from the financial services and manufacturing sectors, where fraud was subsequently confirmed. The AI-supported system demonstrated a 94.7% detection
rate for material misstatements due to fraud, a significant improvement over the
68.2% baseline rate of standard audit procedures applied to the same cases. Furthermore, it reduced false positive alerts by 41% compared to standalone anomaly
detection tools, thereby enhancing audit efficiency. The results substantiate that a
multi-agent, cognitively-inspired AI system can effectively model the tacit knowledge and pattern recognition capabilities of expert forensic accountants, offering a
robust decision-support mechanism. This research contributes a new paradigm for
audit technology, shifting from reactive data mining to proactive, intelligent fraud
hypothesis generation and testing in complex, high-risk environments.